leveraging honest users: stealth command-and-control of botnets

Leveraging Honest Users:

Stealth Command-and-Control of Botnets

Diogo MónicaINESC-ID/IST

[email protected]

mailto:[email protected]

Summary

• Motivation

• Problem statement

• Stealth C&C using browsers

• Final remarks

Motivation

• Botnets continue to evolve

• New strategies must be employed to avoid takedown and detection

• Our objective is to explore new directions future C&C infrastructure might take

Problem Statement

• Create a botnet that:

• Avoids infiltration, size estimation

• Reduces the likelihood of detection of individual bots

• Maintains Botmaster anonymity

Assumptions

• Pre-existing population of infected hosts

• Trust anchor in the binary (public key)

• Bots can receive commands from bot master through some open port

Basic Architecture

• No active participation from bots in a botmaster owned C&C

• Bots passively listen for commands

• Commands are signed by the botmaster and pushed out to all the bots

Basic Architecture

Basic Architecture

• No C&C means:

• no infiltration

• no size estimation

Problems

• Command dissemination

• Botmaster doesn’t know IPs of bots

• Direct dissemination exposes the botmaster

• Disseminating commands takes too long

• Information retrieval

• Bots don’t know the IP of the botmaster

Command Dissemination

• Expendable layer of hosts

• No knowledge about the botmaster

• Do the “heavy lifting” of disseminating commands for the botmaster

Browsers!

• Browsers were created/optimized to do large number of requests per second

• Available crypto libraries in Javascript

• HTML5 brings new capabilities to the table

“Honest” intermediate layer

• Botmaster deploys (or infects) website with malicious code


• Command dissemination is not done by botmaster

• Reduces the vulnerability to detection

• Visitors of the infected website propagate commands

• Dissemination speed increase x #Web Users

• Detecting the existence of a bot is difficult

• Commands are received but not acknowledged


• Replaying the commands will only further spread the botmaster’s orders

• Intermediate layer is expendable and can expire quickly

• Once the page is closed, all traces of “infection” of the web-browser disappear

• It is hard for researchers to find the original malicious page

Analysis of Command Dissemination

• We created Javascript PoC

• Measured the number of AJAX requests per second

• Used EasyXDM to bypass Same-Origin-Policy

• Implemented public-key signatures for commands in Javascript


• N = #bots

• S = #ips in the address space

• r = #requests / second a browser can make

• d = #days the malicious website is active

• v = #visitors per day the website receives

• m = #minutes a user spends on the website


• N = 150000 bots

• S = 3086889768 (2^32 - Bogons)

• r = 250 requests/second

• d = 1day

Getting Visitors

• Create malicious website

• Advertise through spam email, twitter, search engine poisoning, abuse URL shortener, etc

• Infect existing website:

• XSS or SQL injection sufficient to get malicious code on legitimate websites

• Keeping users on the websites

• Tabnabbing, clickjacking

Information Upstream

• Botmasters want to send stolen data upstream (credit-cards, email accounts, SSN’s, etc)

• Our command dissemination infrastructure isolates each bot for robustness and stealthiness, but makes it difficult to create an upstream channel


• For spamming-only botnets a simple solution, send information encoded along with spam

• All information is encrypted with the botmaster’s public key, ensuring confidentiality of data

• The bot only has to do one thing: send spam


• Does not expose the botmaster

• Stealth operation

• Only the botmaster can extract data from the bots


• Botmaster creates website private/public key-pair and signs it with it’s own public key

• The malicious code sent to the browsers includes this key-pair

• Browsers can prove themselves as originating from a “legitimate” dissemination website

Accessing the overlay

Overlay connectivity


• Botmaster randomly scans the internet until it finds one host.

• Uses the encrypted fingers of this host to start crawling through the overlay.

• But...


• Botmaster still needs to bounce through some nodes to guarantee anonymity when retrieving data

Final remarks

• Stealth C&C using browsers are feasible

• Increasing role of browsers in the malware landscape

• We should focus some IDS effort on the browsers

• We aren’t good enough at detecting malicious websites

Thank you

[email protected]

mailto:[email protected]

leveraging honest users: stealth command-and-control of botnets

Technology